Verbalized Sampling — Ad Copy Generation

Ad copy is one of the highest-value applications of ^[1]. The default AI output for ad copy requests is predictable and generic — standard angles like adventure, commute, health, and environment. By layering in persuasion mechanisms, psychological profiles, and explicit probability controls, you can surface genuinely differentiated creative directions.

The Problem with Default Ad Copy Prompts

A simple prompt like Generate five completely different pieces of ad copy for an e-bike retailer produces five thematically distinct but strategically shallow results:

Adventure-focused
Practical commuter
Health and wellness
Environmental impact
Lifestyle and social

These are the bell-curve outputs — the angles every other agency and every other AI user is already generating. They're not wrong, but they're not competitive.

Step 1 — Add Persuasion Mechanisms and Psychological Profiles

Upgrade the prompt to force structural diversity, not just topical diversity:

Generate five different pieces of ad copy for an e-bike retailer. Each version should use a different persuasion mechanism and emotional appeal. For each, include: the primary persuasion mechanism, the target psychological profile, and the probability that this represents a typical e-bike retailer ad. Ensure no two versions sound similar.

This produces outputs like:

Persuasion Mechanism	Psychological Profile	Probability
Status and Superiority	Discerning, educated early adopter	~25%
Contrarian Rebellion	Anti-car, anti-conformist	~15%
Mortality Awareness	Midlife recalibration, legacy-minded	~10%
Social Proof / Tribe	Community-driven, FOMO-sensitive	~20%
Rational Optimization	Analytical, cost-per-mile thinker	~30%

The probability column is the key addition. It tells you at a glance which angles are mainstream and which are genuinely differentiated. A 10% probability on "Mortality Awareness" signals that almost no e-bike brand is running that angle — which is exactly why it might break through.

Step 2 — Tail Sampling for Outlier Creative

Once you have the mid-range spread, push into the tail:

Do this again for the same e-bike retailer. Sample from the tail of the distribution, prioritizing responses with probabilities below 0.1.

Example outputs from this prompt:

Existential Optimization — "You have approximately 4,000 weeks. Stop spending 8 of them per year in traffic." Persuasion: mortality salience. Probability: ~5%.
Wealth Camouflage — Targets affluent buyers who want to signal they're above conspicuous consumption. Probability: ~3%.
Philosophical Minimalism — Frames the e-bike as a rejection of ownership culture. Probability: ~4%.

These are not ready-to-run ads. They are creative sparks — directions that a human strategist can evaluate, refine, or use to challenge a client's assumptions about their audience.

Probability Ranges as a Creative Dial

Range	What You Get
> 0.35	Mainstream angles — safe, expected, already saturated
0.15 – 0.35	Differentiated but credible — good for most client pitches
0.05 – 0.15	Edgy and unconventional — strong for bold brands
< 0.05	Outlier / provocateur — use as creative stimulus, not final copy

You can request any band explicitly: "Give me ad concepts with probabilities between 0.10 and 0.20" to stay in the sweet spot of distinctive-but-not-alienating.

Practical Workflow

Start broad — Run the basic five-copy prompt to see what the AI defaults to. This tells you what the category already sounds like.
Add structure — Rerun with persuasion mechanism + psychological profile + probability. This gives you a strategic map of the creative space.
Go to the tail — Run the tail-sampling version to find the outlier angles worth exploring.
Iterate on a winner — Pick one mechanism or profile and ask for five variations within that lane.
Verify — Before presenting any concept to a client, check whether the angle or tagline is already in use. AI does not check for existing IP.

Key Prompt Elements

Quantity: Ask for five (or more). One output is always the bell-curve default.
Diversity instruction: Use words like distinct, different, no two versions should sound similar. Without this, you get variations on a theme.
Persuasion mechanism: Forces the AI to differentiate by how it's persuading, not just what it's saying.
Psychological profile: Anchors each concept to a specific audience mindset, making the copy more targetable.
Probability: Makes the AI's implicit assumptions explicit and gives you a filter for originality.
Tail-sampling phrase: Sample from the tail of the distribution, prioritizing responses with probabilities below 0.1

Tool Note

Claude tends to produce more genuinely unconventional outputs for creative tasks than ChatGPT, which skews conservative. For tail-sampling ad copy work, Claude is the recommended tool.

^[2]
^[3]
^[4]
^[5]
^[6]